\documentclass{wber}


\usepackage{amsmath,graphicx,array, amssymb, amsthm}
\usepackage{dcolumn,soul}%
\usepackage{amsthm}
\usepackage{lipsum}
\usepackage[figuresright]{rotating}%
\usepackage{algorithm, algorithmicx, algpseudocode}
\usepackage{listings}%
\usepackage[utf8]{inputenc}
\usepackage{booktabs}
\usepackage{adjustbox}

\usepackage[authoryear]{natbib}
\setcitestyle{aysep={,}}

\usepackage{caption}
\bibpunct[, ]{(}{)}{;}{a}{}{,}
\newcommand{\GG}[1]{}
\RequirePackage{url}\def\UrlFont{}
\bibliographystyle{wber}

\makeatletter
%%
\def\uns{\ifmmode\,\else$\,$\fi}%
\newtheorem{theorem}{Theorem}
\newtheorem{construction}{Construction}
\newtheorem{estimate}{Estimate}
\newtheorem{lemma}{Lemma}
\newtheorem{corollary}{Corollary}
\newtheorem{result}{Result}
\newtheorem{algth}{Algorithm}
\newtheorem{proposition}{Proposition}
\newtheorem{hypothesis}{Hypothesis}
\newtheorem{experiment}{Experiment}
\newtheorem{definition}{Definition}
\newtheorem{condition}{Condition}
\newtheorem{property}{Property}
\newtheorem{problem}{Problem}
\newtheorem{fact}{Fact}
\newtheorem{assumption}{Assumption}
\newtheorem{eexample}{Example}
\newtheorem{model}{Model}
%%
\makeatother
 
\raggedbottom

\jvol{00}
\jnum{0}
\jyear{2023}
\doi{10.1093/wber/0000}

\markboth{One, Two, and Three}{One, Two, and Three}


\begin{document}

\dhead{Article}

\title{Heuristics on Call: The Impact of Mobile Phone Based Business Management Advice}

\author{Shawn Cole}

\author{Mukta Joshi}

\author{Antoinette Schoar*}

\affil{Cole: Harvard Business School, scole@hbs.com, Joshi: ideas42, mjoshi@ideas42.org, Schoar: aschoar@mit.edu. We are grateful to Marina Dimova for her contributions to content development. We thank Yuting Wang and Anshul Maudar for excellent research assistance. We are grateful to Janalakshmi, and Negros Women for Tomorrow Foundation (NWTF) for their partnership on this work as well as to Institute for Financial Management and Research (IFMR-LEAD) and Innovations for Poverty Action (IPA) for their support with running the randomized controlled trials in India and the Philippines respectively. Financial assistance from Development Innovation Ventures (DIV) and Consultative Group to Assist the Poor (CGAP) is greatly appreciated.}

\abstract[Abstract]{There is growing evidence that business training for micro-entrepreneurs can be effective. However, in-person training can be expensive and imposes costs on the target beneficiaries. This paper presents the results of a two-site randomized evaluation of a light-touch, mobile phone-based business training service for micro-entrepreneurs in India and the Philippines. The results show that the training had a statistically significant impact on the adoption of improved business practices, with an increase of 0.06 to 0.12 standard deviation points when considering a binary indicator of business practices. The study finds no evidence of impacts on business sales or profits, though the confidence intervals are wide enough to include meaningful effect sizes (positive or negative). These results suggest that mobile phone-based training can be a cost-effective and scalable way to impart business skills to micro-entrepreneurs.}

\jelcode{}

\keywords{}

\maketitle



\section{Introduction}\label{intro}

There are approximately 420 - 510 million micro, small, and medium enterprises (MSMEs) around the world \citep{IFC2013}. Micro, small and medium enterprises account for about 90 percent of businesses and over 50 percent of employment worldwide \citep{ICSM2019} and employ a majority of the population in many low-income countries \citep{IFC2013}. A majority of MSME entrepreneurs do not receive training or support to help them manage the financial complexity of a small enterprise. Yet, it is known that in-person training programs based on  simple business management heuristics or ``rules of thumb" can significantly improve micro-entrepreneurs' business practices and firm revenue in low-income country settings \citep{DrexlerRulesOfThumb}.

This study involves a two-site field experiment whose primary purpose is to examine whether financial heuristics training -- delivered via mobile phone technology -- can affect the management practices of micro-entrepreneurs and improve firm outcomes. The mobile phone-based delivery of this training seeks to overcome key barriers that may hamper the efficacy and reach of ``traditional" business training courses. First, such courses are costly; per-pupil direct cost estimates often range from \$20 to above \$750, and these estimates do not factor in the value of the entrepreneurs' time \citep{McKenzieTrainingDoesntWork}. These programs are often subsidized by governments \citep{ILO2017} and few, if any, charge tuition to beneficiaries, suggesting a low willingness to pay. Nevertheless, developing low-cost, convenient ways to reach MSME owners may be particularly important, as a recent meta-analysis finds such programs can have modest positive effects \citep{McKenzieWBRObs, McKenzieTrainingDoesntWork}.

In contrast to traditional training, distilling information into actionable, simple rules of thumb may lower adoption barriers and improve financial management even in the absence of complete understanding of accounting or business planning \citep{FeldmanSimplicity, MaddoxCognition}. Financial heuristics or rules of thumb  which lighten the cognitive burden of learning useful financial management skills may therefore be more effective for many micro-entrepreneurs in low-income countries who face frequent scarcity of money and time \citep{ms2013}. \cite{DrexlerRulesOfThumb}, in a field experiment in the Dominican Republic, demonstrate that in-person classroom training of this sort can improve micro-entrepreneurs' management practices and revenues. \cite{BhanotLessIsMore} use a randomized controlled trial to test the effects of a traditional training program relative to the effects of a tailor-made heuristics-based program for micro-entrepreneurs in Ecuador and find statistically and economically meaningful incremental effect sizes on sales and profit of the tailored-heuristics training.

Building on \cite{DrexlerRulesOfThumb}, this intervention delivers and evaluates a similar financial heuristics training via lower-touch mobile phone delivery. A key advantage of mobile delivery is that programs can be scaled quickly, at low cost, with  high fidelity (a recorded message is identical whether sent to one thousand or one million people, while scaling traditional ``train the trainers'' programs could require significant human resources and management). Partnering with microfinance institutions (MFIs) in a two-site field experiment in the Philippines and India, financial heuristics training content was delivered to micro-entrepreneurs via weekly audio messages over their mobile phones. Combining various recommended practices into a binary indicator of improved practices, this study finds that the low-touch intervention increases adoption of recommended practices (on a binary scale) by 0.01 to 0.019 (0.06 to 0.12 standard deviation points), roughly 20-40 percent of the effect size found in \cite{McKenzieTrainingDoesntWork}'s meta-analysis of higher-cost and higher-touch traditional classroom training programs. While evidence of change in business practices is seen, effects on sales and profits are substantially noisier with point estimates of effects statistically indistinguishable from zero. 

This paper contributes to two broad strands of related literature. The first is on managerial skill, firm productivity and economic growth. Managerial skill can contribute to firm productivity \citep{AntoinetteMissingCapital}. Some businesses in low-income country settings employ poor business practices \citep{BloomManagement} and these practices hamper productivity \citep{HsiehMisallocationQJE}. In a randomized controlled trial with Peruvian group lending clients, \cite{karlan2011} find little or no evidence of any marginal effect of an add-on business training module on revenues and profits.  \cite{McKenzieTrainingDoesntWork} provides a comprehensive review of the training literature. While some individual studies find that customized management consulting advice can improve management practices and firm outcomes \citep{BloomJEP, AntoinetteConsultingJPE, Dalla2021}, others are underpowered to detect meaningful effects on sales and profits. \cite{McKenzieTrainingDoesntWork}'s meta-analysis finds positive effects on profits and sales in the order of magnitude of 5 to 10 percentage points.

Relative to the literature, this study finds that a much lower-touch intervention- mobile-phone based heuristic financial advice- delivered over a longer period of time, in the preferred specification, can have 20-40 percent of the effect of these much higher-touch interventions. Additionally, the intervention is longer lasting and less affected by the attendance and attrition problems other studies have faced.\footnote{Many traditional training programs deliver training over multiple days, typically a week, and require participants to travel to a training site. In this setting participants do not need to travel to obtain training contributing to a low training attrition rate. In addition, it can be conjectured that the borrowing relationships between participants and partner microfinance institutions (through whom this training was delivered) may have also led to the low levels of attrition.} It is, additionally, in sites where few papers in this literature have reported on.\footnote{A close review of the literature found no studies set in the Philippines, and only one set in India, which evaluates the impact of business training on micro-entrepreneurs. \cite{Dalla2021} find management training offered by a microfinance organization in India has a positive effect on financial management skills. \cite{BloomManagement} was conducted in India but involves an intense management consulting program.}

The work presented in this paper is perhaps most closely related to \cite{Acimovic2020}, which reports on a field experiment conducted by a mobile network operator with a goal of encouraging its independent sales agents to improve their inventory management. Agents who received both in-person training and explicit recommendations improved their performance.

Lastly, this paper contributes to the literature on digital service delivery in low-income countries. \cite{ColeAgAdoption} find that agricultural advice delivered via a phone-based platform improves farmer decision making, and a meta-analysis by \cite{KremerScience} suggests that such approaches are effective at increasing agricultural yield in a variety of contexts. \cite{AkerAEJLearning} find that basic skills can be taught via mobile phone. The present study extends the application of mobile phone-based delivery by  showing that a mobile phone-based heuristics training can in fact change entrepreneurs’ business management behaviors. 


\section{Setting: Training and Intervention}\label{setting}

This study was conducted with micro-entrepreneurs in India and the Philippines. Micro, small and medium enterprises employ a substantial share of the national labor force in both countries: MSMEs are 99\% of all registered enterprises in the Philippines \citep{CPBRD2020} and employ approximately 63 million workers in India \citep{MMSMEI(2019)}. Additionally, both countries have high mobile phone penetration \citep{ITC2019}.

In both countries, the mobile phone-based heuristics training was rolled out to a subset of MSME clients of two trusted local MFIs. In the Philippines, the team partnered with Negros Women for Tomorrow Foundation (NWTF), while in India, Janalakshmi was the partner MFI. In collaboration with these partners, extensive qualitative interviews with micro-entrepreneurs were conducted to understand the bottlenecks their businesses faced, the kinds of training that would be useful, and different delivery alternatives that could be explored. Prior to this study, a robust pilot was run in India (with Janalakshmi and Institute for Financial Management and Research LEAD) which differed materially in execution, but was similar in spirit. That pilot found no effects on behavior or business outcomes. Refinements to that pilot - in content, delivery method, technology partners, etc. – in conjunction with participating micro-entrepreneurs’ interviews informed the structure of this intervention. The training curriculum was developed by a partner non-profit, ideas42, that oversaw the overall execution and project management of the field experiment. They did so in concert with extensive field interviews with partner institutions and micro-entrepreneurs and learnings from the pilot study. 


Treated entrepreneurs received a 30-minute in-person orientation to the training that they would receive over the subsequent 20+ weeks via pre-recorded weekly interactive voice response (IVR) messages on their mobile devices. The in-person orientation sessions were held in group settings with 25-30 treated entrepreneurs in each session.\footnote{Clients who were unable to attend the in-person training were oriented individually.} The orientation session had three objectives. First, they introduced the training program to the treated MFI clients and set their expectations on what to expect over the coming weeks. Second, it ensured the MFI clients knew how to use their mobile phones to receive incoming IVR calls.\footnote{The IVR service was based on outgoing ``push” calls, which automatically called MFI clients and played a recorded message when MFI clients answered the phone. Treated clients could also dial the training phone number and leave a missed call in order to trigger a call back to listen to the previous two weeks’ messages. These outbound weekly calls were free for treated MFI clients. The decision to use voice calls meant that all mobile phone types were compatible with the service (most of the micro-entrepreneurs in the study had basic phones, and not smart phones), and sidestepped concerns about literacy with text-based training.} Finally, the in-person orientation introduced clients to the concept of “cash separation”, one of the four key pillars of the training curriculum.\footnote{The MFI partners felt in-person orientation was necessary due to the novel nature of mobile-phone service delivery. The overall treatment can be thought of as a bundle of financial heuristics delivered via mobile phone and this brief in-person orientation. The majority of treated clients received the orientation: 95\% in the Philippines and 92\% in India.} Clients were introduced to the first module at the brief in-person orientation session where they were given two handouts in their local language. One summarized how to access the training service via their phones and the second was a visual aid describing the concept of cash separation between household and business. 

The training curriculum was built off of \cite{DrexlerRulesOfThumb} and contained simple rules of thumb on business management organized into four modules: 
\begin{enumerate}
    \item \textit{Cash Separation (Profit Calculation):} Presented micro-entrepreneurs with simple action steps around how to separate business and household cash, and to pay themselves a fixed weekly salary, in order to better monitor their own business’ profitability.
    \item \textit{Customer Credit:} Provided simple rules of thumb on when and when not to offer credit to customers.
    \item \textit{Inventory Management:} Presented simple action steps on how to manage inventory of a retail business. 
    \item \textit{Supplier Management:} Provided action steps on selecting reliable suppliers that offer the best price and product quality.
\end{enumerate} 
Exemplary messages of the specific heuristics taught are provided in Online Appendix 1. The core of the intervention was 3-4 minute long audio messages that delivered the training content each week via IVR calls to treated micro-entrepreneurs. Entrepreneurs were called at their preferred times, which they had specified at the in-person orientation. Entrepreneurs could also access a missed call service by ringing the training number to receive a free call back to access the training content from the current week and the previous week. The training was offered free of charge to participants. 
 

The training was delivered in a soap opera format to bolster engagement. The lead character (female voice actor) in the series played the part of a successful small business owner with years of experience who offered the micro-entrepreneurs practical tips on business management that she had learned over the years from running her own business. The messages were delivered to the treated entrepreneur in their local language weekly for a total of 21 weeks in the Philippines (between August 2016 and January 2017) and 22 weeks in India (between August 2016 and March 2017).\footnote{The India sample was augmented after the start of the experiment in a manner explained in Section \ref{design}.} \footnote{In the Philippines, 21 messages were delivered as part of the heuristics training on availing cash discounts from suppliers was not relevant for the Filipino entrepreneurs who buy all their stock in cash.} The per participant cost of training delivery (airtime only) in India was \$2.04 for messages in Hindi (total of 73 minutes) and \$2.38 for messages in Kannada (total of 85 minutes). In the Philippines, where airtime charges are much higher, the per participant cost of training delivery was \$14.99 (total of 81 minutes).\footnote{This per participant cost includes the airtime used for listening to all training messages in entirety. Actual costs incurred are in fact lower, as the service only charges for airtime used in a call. Because content from a successful intervention could quite easily be scaled to reach hundreds of thousands (or even millions) of entrepreneurs, focus is put on marginal costs, rather than the content development cost. The primary content development costs were staff time of ideas42.} 

In this study, microfinance institutions, which have been extensively studied, were partnered with. \cite{Banerjee2015}, as well as \cite{Meager2022a} summarize the evidence of the causal effects of expansion of microcredit, and, importantly, find heterogeneous effects, with limited impact on sales or profits for the median borrower, but large gains for borrowers at the upper tail. This study seeks to measure the incremental benefit of advisory services on top of credit.

\section{Experimental Design}\label{design}

The two-site experiment was conducted in the Philippines and in India. The experimental design in each setting is described below.

The field experiment in the Philippines ran from March 2016 to June 2017. The sample was drawn from active group loan clients of the partner MFI - NWTF. To be eligible for the experiment clients had to speak Hiligaynon, manage a retail business and have access to a mobile phone. The baseline data collection exercise was conducted from March to June 2016. It consisted of in-person interviews with eligible clients to gather detailed information on their demographics, business ownership, financial and managerial practices, and business outcomes. A total of 2,096 clients were interviewed. Clients were then randomly assigned into a control arm (1,030 clients) and a treatment arm (1,066 clients). As the clients were beneficiaries of group loans, randomization was carried out at the group level to account for the possibility of spillovers; 676 groups were randomly assigned to treatment and 675 into control. Stratified group randomization by the number of members in each group (which ranged from one to five) was done to ensure an equal number of treatment and control clients for each stratum of group size.

Clients in the treatment group received the in-person orientation described in section 2. These sessions were conducted by NWTF staff and held at weekly MFI group meeting sites where group members met on a weekly basis as part of their group lending program with NWTF. About 78 percent of treatment clients attended the group orientation sessions. For those clients who were unable to attend the initial large group session,  make-up orientation was held by NWTF staff in a smaller group or individual settings. Roughly 5 percent of treatment clients did not receive any orientation. Treatment group clients received weekly heuristic training messages for a total of 21 weeks between August 2016 and January 2017s. Endline in-person surveys were conducted between April and June 2017. A total of 1,898 clients were surveyed at the endline out of the baseline sample of 2,096.

The field experiment in India ran from March 2016 to July 2017. The sample was drawn from individual loan clients of partner MFI Janalakshmi. To be eligible for being part of the sample, clients had to speak Kannada, Hindi or Urdu, and - just as in the Philippines - manage a retail business and have access to a mobile phone. 
The study began with a total sample of 2,407 clients in Bangalore (henceforth referred to as Wave-1). The sample was increased by 1,442 clients (Wave-2) midway through the experiment in order to increase the study’s statistical power. The Wave-2 sample was drawn based on the same eligibility criteria and included clients from Bangalore, Mysore, Davangere, Gulbarga, Indore, and Delhi. Baseline data collection for Wave-1 clients was conducted between March and May of 2016. However, a baseline survey for Wave-2 clients was not conducted. In India randomization was conducted at the individual level. For Wave-1 clients, randomization was stratified by MFI branch in order to ensure an equal number of treatment and control clients at each branch. Wave-1 randomization was conducted after baseline data collection. Wave-2 randomization was also conducted at the individual level but stratified by region and language to ensure an equal proportion of treatment and control clients in each region and for each language. The in-person, group orientation sessions were held at Janalakshmi’s local offices. Only 22 percent of clients attended these sessions. Non-attendees were followed-up individually for a make-up orientation conducted on an individual basis by the research team.\footnote{Innovations for Poverty Action (IPA) and Institute for Financial Management and Research (IFMR-LEAD) oversaw the implementation of the experiment in the Philippines and India respectively.} About 70 percent of treatment clients received individual make-up orientation sessions. Heuristic training messages were sent out for a total of 22 weeks in India. Wave-1 clients received messages from August to December 2016 and were interviewed for endline data collection between April and June 2017. Wave-2 clients received messages from October 2016 to March 2017. Their endline surveys were conducted between June and July 2017. A total of 3,318 clients were surveyed at the endline out of the baseline sample of 3,849.


\section{Data and Estimation Strategy}\label{estStrat}

\subsection{\textbf{Data}}

The primary data sources of this study are the in-person interviews conducted at baseline and endline with the study participants. The baseline and endline surveys in both sites were conducted in private, one-on-one interviews between enumerators and the entrepreneurs, who were informed that the financial institution would not have access to their individual data from the surveys. For impact evaluation, self-reported measures, including those on firm productivity, which are imperfect \citep{McKenzieMeasuringProfits} but allow for comparability to a number of other studies, are relied on. These survey responses were augmented with administrative data on the study participants from the partner MFIs. Pick-up and listenership data from the IVR platform provider was also collected to gauge engagement of the treated clients with the training content. 


These survey responses are augmented with administrative data about the entrepreneurs in the sample from the partner MFIs. Pick-up and listenership data from the IVR platform provider was also collected to gauge engagement of the treated clients with the training content.

Tables 1A \& 1B report baseline summary statistics and balance tests for the two samples. In the Philippines, the typical study participant is, on average, a 45 year-old female entrepreneur, who in 72 percent of cases has a high school diploma. In the Philippines, 53 percent of the sample entrepreneurs report that their current business is their primary source of income. The sample in India is somewhat different. The typical study participant is, on average, a 38 year-old female entrepreneur. Only 35 percent of the Indian sample has a high-school diploma and 49 percent report that their business is their primary source of income. 

The second panel reports a snapshot of the business practices sampled entrepreneurs adopt at the baseline. In the Philippines, 68 percent report they separate business and household cash, 77 percent calculate profits, 78 percent give credit for no more than seven days, 77 percent keep business records, 70 percent keep records of customer credit, 68 percent record important information of customer credit, and 76 percent take full advantage of cash discounts offered by suppliers. These baseline adoption measures are similar to those in \cite{DrexlerRulesOfThumb} in the Dominican Republic wherein 74 percent of sampled entrepreneurs reported separating business and personal cash, 66 percent kept reports of their accounts and 81 percent formally calculated their revenues. 

Baseline adoption of recommended business practices is considerably lower in India, as Table 1B illustrates. Of the sampled entrepreneurs at baseline (Wave-1), only 35 percent separate household and business cash, 81 percent calculate profits, 28 percent give credit to customers for a week and only 40 percent keep business records. Unlike in the Philippines, only 42 percent keep records of customer credit, 10 percent note important customer credit details and 39 percent take advantage of cash discounts offered by suppliers.


Average sales for sampled entrepreneurs are about 6,094 (\$128) pesos in the Philippines and 14,115 rupees  (\$210) in India. Profits in a regular week are, on average, 2,344 pesos (\$49) in the Philippines and about 5,233 rupees (\$78) in India. In  \cite{DrexlerRulesOfThumb} average week sales were on average 6,399 Dominican Pesos (\$181).\footnote{Exchange rate in 2016: India: USD 1=INR 67.18; The Philippines: USD 1= PHP 47.49. The intervention in \cite{DrexlerRulesOfThumb} was implemented in the Dominican Republic between March and May 2007 at which time the exchange rate was roughly 1 USD = 35.29 Dominican Pesos. Exchange rates sourced from: https://www.exchangerates.org.uk/.} An important strength of this study is that it tests whether similar business training can affect business practices in three different settings - starting with the Dominican Republic and then in India and the Philippines. 

In comparing the two study sites, it is worth noting that the baseline adoption of recommended practices is substantially higher in the Philippines. This may be due to the higher level of education of the Filipino entrepreneurs in the sample (72 percent high school or better versus 35 percent) since education level and business practices are positively correlated (correlation not reported). Additionally, the market structures likely vary substantially across the two sites. 

Listenership rates in India and the Philippines are reported in Tables 2A and 2B. These rates provide a snapshot of participant engagement with the program. The average pickup rate across the two countries was 76 percent. Listenership rate conditional on pickup across the two countries was 84 percent.


\subsection{\textbf{Estimation Strategy}}

The primary purpose of this paper was to examine whether mobile phone-based business training could impact business practices and firm productivity. As described earlier, this study was inspired by \cite{DrexlerRulesOfThumb}, and thus focuses on the outcomes identified in that paper, namely, business practices, sales, and profits.\footnote{A pre-analysis plan was not created.} Since treatment was randomly assigned, the differences estimator, specified in the model below, provides unbiased estimates of the target estimand -- the training's average treatment effect: 

\begin{equation}\label{model1}
    y_i^E = \alpha + \beta_1 T_i + \beta_2 y_i^B + \gamma W_i + \varepsilon_i 
\end{equation}
where $y_i^E$ is the endline outcome of interest, $T_i$ is a treatment dummy, $W_i$ is a vector of controls and $y_i^B$ is the baseline measure of the outcome variable and is included where available.\footnote{Covariates in the Philippines include location, age of business, own a cellphone indicator, primary source of income indicator, education level, business type, and variables used for stratification: number of clients in each group. Covariates in India include wave dummy, time of survey, gender, age of business, own a cellphone indicator, primary source of income indicator, education level, business type, and variables used for stratification: branch and language.}

The outcomes of interest are business practices, sales and profits. The business practice measures collected at endline were enumerated on a three-point scale, with 1 being the least desired and 3 being the most desired outcome in reference to the practices taught in the training. 

For instance, it was asked how often entrepreneurs contact customers whose credit is due. If an entrepreneur said “None of the time” that would get a score of 1, “Some of the time” that would get a score of 2, and “Often / all of the time” that would get a score of 3.  Much of the literature reports business outcomes as binary practices, rather than the ranges used in this paper. For example, \cite{McKenzieMeasuringProfits} state “for every 20 practices that business training attempts to teach firms to do, on average firms invited to training only implement one additional practice.” To provide some (admittedly imperfect) comparability, for each business practice, this paper converts the three-value range into a binary variable.\footnote{This is done by assigning the low value 0, the high value 1, and the intermediate value to zero if there are more lows than highs for that item, or to 1 if there are more highs than lows. This, in effect, passes the three-point scale through an above/below median filter to convert it into a binary outcome.}$^{,}$\footnote{The relevant business practices include separating business and household cash, paying a fixed weekly salary to self, calculating profits, giving credits for no more than seven days, calling the customers whose credit is due, keeping business and credit records, buying more of the most popular products and less of the least popular products, visiting competitors to check out price, talking to customers to check out need, introducing new product, comparing price and quality of various suppliers, negotiating prices and terms with suppliers, and taking advantage of cash discount from suppliers.} Both results are presented in the outcome table, but the binary index measure is discussed for comparability to other papers.

In terms of productivity measures, ``regular week" sales and profits are considered as outcomes of interest. These are reported in levels winzorised at the 1 percent level as well as in logs.\footnote{It can be noted regarding the log transformation that no entrepreneurs in the sample reported null or negative profits for a ``regular week”.} For both sales and profits, respondents were asked three questions at endline. The first two included enumeration of sales / profits on the previous day and an assessment of whether the previous day’s profits were ‘good’, ‘bad’ or ‘regular’. This was followed by respondents being asked to report sales / profits in a typical week.\footnote{The precise wording of the sales question was: ``(1) What were the sales in your business yesterday? (Sales from primary business only); (2) How would you classify yesterday in terms of sales?; (3) Can you tell us what the average sales per week are in your business?” Respondents were asked the same question for profits.}

Under this baseline model standard errors are clustered at the group level in the Philippines given the weekly-group meetings in that setting. Heteroskedasticity robust standard errors are reported in India where randomization was done at the individual level.

The analysis also tests for heterogenous treatment effects along four dimensions of heterogeneity, namely, the entrepreneur’s level of education, age, business size and baseline adoption of recommended business practices. This is done by running the following model on the full sample. 
\begin{equation}\label{model2}
    y_i^E = \alpha + \beta_1 T_i + \beta_2 X_i + \beta_3 T_i*X_i + \beta_4 y_i^B + \gamma W_i +  \varepsilon_i 
\end{equation}
where $X_i$ is a dummy that is turned on when the entrepreneur has an above median measure of the relevant axis of heterogeneity being tested. For instance, for age $X_i = 1$  if the entrepreneur is as old or above the median age of the sample and zero otherwise. In this model, $\hat{\beta}_3$ is the estimate of interest. To correct for multiple hypothesis testing in the heterogeneity analysis, we calculate and report \cite{AndersonQvalues} sharpened q-values.

\section{Results}\label{results}

\subsection{\textbf{Uptake \& Engagement}}

Take-up and engagement with the training content is presented in Tables 2A \& 2B. In India, the mean (median) number of calls answered was 16.7 (19), while in the Philippines it was 15.3 (18). The uptake of the program, as measured by pick-up rates, was high - with around three quarters of calls picked up across all four training modules in both sites. Moreover, participants were engaged with the training as listenership rates were above 60 percent in the Philippines and about 48 percent in India\footnote{An individual’s listenership rate is calculated as the total number of minutes listened to by a client (across all messages) divided by the total number of minutes of content the participant would have heard had they listened to all messages in entirety. If a client never answered a call, then her listenership rate would be zero.}.  Participants were more engaged with the first two modules, which covered cash separation and customer credit, compared to the later modules on inventory and supplier management. Additionally, around 32 percent of treated beneficiaries in the Philippines used the missed call service at least once and about 40 percent did so in India.

As part of the endline data collection, participants’ feedback on the training was also collected. About 78 percent of the training participants in the Philippines and 62 percent of the training participants in India reported that they were likely to recommend this training program to their family, friends, and other business owners like them.

Tables 3A \& 3B report predictors of engagement with the training content. Pick-up and listenership rates increase with age in the Philippines but not in India. Owning a cellphone (as opposed to having access via some other means, say a family member, for instance) is the strongest predictor of engagement with the training calls in both contexts. In the Philippines, education is negatively correlated with engagement. Engagement does not appear to vary materially by other covariates.

\subsection{\textbf{Impact on Practices and Productivity}}

The main experimental findings are reported in Table 4. Financial heuristics training delivered via mobile phones increases the adoption of recommended business practices by 0.01 (India)-0.02 (Philippines) on the unweighted average binary business practice index. These results are significant at the 10 percent and 5 percent levels respectively. The magnitude of this change in business practices is 0.06-0.12 standard deviation points, an effect size of 20-40 percent of the magnitude of the effect of traditional training programs suggested by  \cite{McKenzieTrainingDoesntWork}'s meta-analysis. The results are of comparable magnitude but more statistically significant when expressed on the three-point range over which initial survey responses were enumerated.

No statistically significant changes are found in firm productivity. Point estimates are positive for revenues in the Indian sample and mildly negative in the Philippines. Point estimates for profits are near zero for both samples as well. For both sales and profits in both sites none of the productivity measures are statistically distinguishable from zero at the 10 percent confidence level.

Attrition is relatively low in both the Philippines (treatment (9.1 percent) and control (9.8 percent) and India (treatment (13.88 percent) and control (13.71 percent)). In both settings the attrition rate is statistically indistinguishable across treatment and control groups. Evidence of differential attrition is examined by testing whether there is a statistically significant difference in baseline variables between the treatment and control observations in the attrited samples. However, no statistically significant difference is found (Online Appendix Tables 2-3).

Why might not the same positive, statistically significant effects on sales and profits in this sample as \cite{DrexlerRulesOfThumb} report in the Dominican Republic be observed? A number of possibilities are explored. One possibility is that mobile phone-based intervention is simply weaker: while the barriers to attendance were lower, the total ``contact time” in this intervention was much lower.\footnote{An alternative possibility is that entrepreneurs in the Dominican Republic respond more to business training; unfortunately, conducting this study with a sufficient number of entrepreneurs in the Dominican Republic would have been prohibitively expensive.} Another possible explanation is that generalized heuristics are not optimized for specific entrepreneurs and as such will not drive growth in sales and profits in some specific settings for some entrepreneurs. Participants in India were affected by a demonetization policy that may have negatively impacted business. Moreover, there is in fact very little systematic evidence isolating the \textit{specific} business practices that drive the most sales and profit growth. Like most of the literature, this paper evaluates a bundled program, based on the best curriculum that could be devised for this study. But perhaps the suggestion of credit limits for certain customers in some contexts, for example, might have limited sales.

To summarize the main finding, the primary purpose of this study was to examine whether mobile phone-based training could affect business practices. While evidence of change in business practices is seen, a shortcoming of this paper's approach is that there is limited information on financial performance. While the business practices are relatively precisely measured, financial outcomes such as sales and profits, are unfortunately substantially noisier. Even with a sample of 3,849 entrepreneurs in India and 2,096 in the Philippines, when we consider regular week profits winsorized at the 1 percent level, the standard error is  approximately 5 percent of the sample mean in the Philippines and about 3.4 percent in India, which prevents the ruling out of meaningful economic effects.

Like most evaluations in this literature, only one program is evaluated, and therefore the paper is unable to answer important questions such as whether training combined with credit is more effective than training alone, or whether the identity of the provider of information affects take-up. These are important design questions, and it can be noted that a digitally designed and delivered service, such as the one evaluated here, may be particularly well suited to investigate these questions, through for example a series of A/B tests with a large population.


\subsection{\textbf{Heterogeneity Analysis}}

There is no reason to believe that treatment effects must be homogenous, and the differential findings in the Dominican Republic, India, and the Philippines suggest it is worth exploring treatment heterogeneity. Tables 5 \& 6 report heterogeneous effects, as estimated by Equation (\ref{model2}), along the dimensions expected to matter most.

Four dimensions of heterogeneity are focused on, namely, the entrepreneur’s level of education, age, business size, and baseline adoption of recommended business practices.  Differences along the education dimension help to assess if the effectiveness of training depends on the baseline level of human capital where we predict that micro-entrepreneurs with lower levels of educational attainment might be able to understand and apply the practices equally well as micro-entrepreneurs with higher levels of educational attainment given that the rule of thumb training is designed to be relatively easy to implement regardless of one's educational background. The hypothesis that older individuals may be ``set in their ways” is analyzed by measuring treatment heterogeneity by age. 

Testing along business size (baseline regular week sales used as a proxy) helps examine if there are differential treatment effects for small versus large businesses. Micro-entrepreneurs with larger businesses may be less willing to change practices that have enabled them to reap larger sales. Additionally, micro-entrepreneurs with lower sales may have a greater incentive to learn and adopt more effective practices to increase their sales. 

Finally, treatment heterogeneity by baseline business practice adoption is also tested. It is hypothesized that micro-entrepreneurs with lower baseline adoption of recommended financial practices may have larger room for improvement and thereby see greater gains from the training.

As Tables 5 \& 6 illustrate, there is evidence of heterogeneous treatment effects in the Philippines. Specifically, it is seen that the treatment to be twice as effective among young entrepreneurs compared to older entrepreneurs, and to be substantially more effective among small businesses. 

The results in the Philippines suggest the training is more effective at changing the behavior of younger entrepreneurs who are likely less set in their business practices. This is consistent with the effect size being larger for small businesses as well, as they might have greater flexibility to change and adopt new practices. 

In India, no evidence of heterogeneity along the four dimensions is found. 



\section{\textbf{Conclusion}}\label{conclusion}

This paper presents the results of a two-site randomized experiment assessing the impact of mobile phone based financial heuristics training on micro-entrepreneur's business practices and firm outcomes. The training intervention was taken up by most entrepreneurs and the majority engaged with the content as illustrated in Table 2. It is found that the training led to a significant improvement in business practices amongst the treated micro-entrepreneurs. The effect size estimate on improved business practices ranges between 0.06 and 0.12 standard deviation points of the practice adoption index. In the wider literature that focuses on much higher-touch interventions, effect sizes of the order of magnitude of 20-40 percent are found.

The paper extends the work in the training and firm productivity literature. The focus on a mobile phone-based intervention allows for a greater scalability due to lower implementation costs. The more modest effect sizes might be more beneficial on a cost-benefit basis since the marginal cost of extending this training to other entrepreneurs is negligible. 


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\begin{thebibliography}{}

\bibitem[\protect\citeauthoryear{Acimovic, Parker, Drake, and
  Balasubramanian}{Acimovic et~al.}{2020}]{Acimovic2020}
Acimovic, J., C.~Parker, D.~Drake, and K.~Balasubramanian. 2020.
\newblock ``Show or tell? improving inventory support for agent-based
  businesses at the base of the pyramid.''
\newblock

\bibitem[\protect\citeauthoryear{Aker, Ksoll, and Lybbert}{Aker
  et~al.}{2012}]{AkerAEJLearning}
Aker, J.~C., C.~Ksoll, and T.~J. Lybbert. 2012, July.
\newblock ``Can mobile phones improve learning? evidence from a field
  experiment in niger.''
\newblock {\em American Economic Journal: Applied Economics\/}~{ 4\/}~(4):
  94--120.

\bibitem[\protect\citeauthoryear{Anderson}{Anderson}{2008}]{AndersonQvalues}
Anderson, M.~L. 2008.
\newblock ``Multiple inference and gender differences in the effects of early
  intervention: A reevaluation of the abecedarian, perry preschool, and early
  training projects.''
\newblock {\em Journal of the American Statistical Association\/}~{
  103\/}~(484): 1481--1495.

\bibitem[\protect\citeauthoryear{Arráiz, Bhanot, and Calero}{Arráiz
  et~al.}{2019}]{BhanotLessIsMore}
Arráiz, I., S.~P. Bhanot, and C.~Calero. 2019.
\newblock ``Less is more: Experimental evidence on heuristic-based business
  training in ecuador.''
\newblock {\em DB Invest Working Paper TN\/}~{18}.

\bibitem[\protect\citeauthoryear{Banerjee, Karlan, and Zinman}{Banerjee
  et~al.}{2015}]{Banerjee2015}
Banerjee, A., D.~Karlan, and J.~Zinman. 2015, January.
\newblock ``Six randomized evaluations of microcredit: Introduction and further
  steps.''
\newblock {\em American Economic Journal: Applied Economics\/}~{ 7\/}~(1):
  1--21.

\bibitem[\protect\citeauthoryear{Bloom, Eifert, Mahajan, McKenzie, and
  Roberts}{Bloom et~al.}{2012}]{BloomManagement}
Bloom, N., B.~Eifert, A.~Mahajan, D.~McKenzie, and J.~Roberts. 2012, 11.
\newblock ``{ Does Management Matter? Evidence from India}.''
\newblock {\em The Quarterly Journal of Economics\/}~{ 128\/}~(1): 1--51.

\bibitem[\protect\citeauthoryear{Bloom and Van~Reenen}{Bloom and
  Van~Reenen}{2010}]{BloomJEP}
Bloom, N., and J.~Van~Reenen. 2010, March.
\newblock ``Why do management practices differ across firms and countries?.''
\newblock {\em Journal of Economic Perspectives\/}~{ 24\/}~(1): 203--24.

\bibitem[\protect\citeauthoryear{Bruhn, Karlan, and Schoar}{Bruhn
  et~al.}{2018}]{AntoinetteConsultingJPE}
Bruhn, M., D.~Karlan, and A.~Schoar. 2018.
\newblock ``The impact of consulting services on small and medium enterprises:
  Evidence from a randomized trial in mexico.''
\newblock {\em Journal of Political Economy\/}~{ 126\/}~(2): 635--687.

\bibitem[\protect\citeauthoryear{Bruhn, Schoar, and Karlan}{Bruhn
  et~al.}{2010}]{AntoinetteMissingCapital}
Bruhn, M., A.~Schoar, and D.~Karlan. 2010, May.
\newblock ``What capital is missing in developing countries?.''
\newblock {\em American Economic Review\/}~{ 100\/}~(2): 629--33.

\bibitem[\protect\citeauthoryear{Cole and Fernando}{Cole and
  Fernando}{2020}]{ColeAgAdoption}
Cole, S.~A., and A.~N. Fernando. 2020, 07.
\newblock ``{‘Mobile’izing Agricultural Advice Technology Adoption
  Diffusion and Sustainability}.''
\newblock {\em The Economic Journal\/}~{ 131\/}~(633): 192--219.

\bibitem[\protect\citeauthoryear{CPBRD}{Congressional Policy and Budget Research Department, Congress of the Philippines}{2020}]{CPBRD2020}
CPBRD. 2020.
\newblock ``Msmes in the philippines.''.
\newblock Congressional Policy and Budget Research Department - Congress of the
  Philippines.

\bibitem[\protect\citeauthoryear{dalla Pellegrina, Skully, Walker, Wanke, and
  Wijesiri}{dalla Pellegrina et~al.}{2021}]{Dalla2021}
dalla Pellegrina, L., M.~Skully, T.~Walker, P.~Wanke, and M.~Wijesiri. 2021,
  October.
\newblock ``Does ownership structure affect firm performance? evidence of
  indian bank efficiency before and after the global financial crisis.''
\newblock {\em American Economic Journal: Applied Economics\/}~{ 29\/}~(3):
  1842--1867.

\bibitem[\protect\citeauthoryear{{de Mel}, McKenzie, and Woodruff}{{de Mel}
  et~al.}{2009}]{McKenzieMeasuringProfits}
{de Mel}, S., D.~J. McKenzie, and C.~Woodruff. 2009.
\newblock ``Measuring microenterprise profits: Must we ask how the sausage is
  made?.''
\newblock {\em Journal of Development Economics\/}~{ 88\/}~(1): 19--31.

\bibitem[\protect\citeauthoryear{Drexler, Fischer, and Schoar}{Drexler
  et~al.}{2014}]{DrexlerRulesOfThumb}
Drexler, A., G.~Fischer, and A.~Schoar. 2014, April.
\newblock ``Keeping it simple: Financial literacy and rules of thumb.''
\newblock {\em American Economic Journal: Applied Economics\/}~{ 6\/}~(2):
  1--31.

\bibitem[\protect\citeauthoryear{Drexler, Fischer, and Schoar}{Drexler
  et~al.}{2014}]{DrexlerRulesOfThumb}
Drexler, A., G.~Fischer, and A.~Schoar. 2014, April.
\newblock ``Keeping it simple: Financial literacy and rules of thumb.''
\newblock {\em American Economic Journal: Applied Economics\/}~{ 6\/}~(2):
  1--31.

\bibitem[\protect\citeauthoryear{Fabregas, Kremer, and Schilbach}{Fabregas
  et~al.}{2019}]{KremerScience}
Fabregas, R., M.~Kremer, and F.~Schilbach. 2019.
\newblock ``Realizing the potential of digital development: The case of
  agricultural advice.''
\newblock {\em Science\/}~{ 366\/}~(6471): eaay3038.

\bibitem[\protect\citeauthoryear{Feldman}{Feldman}{2003}]{FeldmanSimplicity}
Feldman, J. 2003.
\newblock ``The simplicity principle in human concept learning.''
\newblock {\em Current Directions in Psychological Science\/}~{ 12\/}~(6):
  227--232.

\bibitem[\protect\citeauthoryear{Hsieh and Klenow}{Hsieh and
  Klenow}{2009}]{HsiehMisallocationQJE}
Hsieh, C.-T., and P.~J. Klenow. 2009, 11.
\newblock ``{Misallocation and Manufacturing TFP in China and India*}.''
\newblock {\em The Quarterly Journal of Economics\/}~{ 124\/}~(4): 1403--1448.

\bibitem[\protect\citeauthoryear{International Council~for
  Small~Business}{International Council~for Small~Business}{2019}]{ICSM2019}
International Council~for Small~Business, I. 2019.
\newblock ``Annual global micro, small and medium-sized enterprises report.''.

\bibitem[\protect\citeauthoryear{{International Finance Corporation,}}{{International Finance Corporation}}{2013}]{IFC2013}
{International Finance Corporation}. {2013}.
\newblock ``{Small and Medium Enterprise Finance: New Findings, Trends and
  G-20/Global Partnership for Financial Inclusion Progress}.''.

\bibitem[\protect\citeauthoryear{International
  Telecommunication~Union}{International
  Telecommunication~Union}{2019}]{ITC2019}
International Telecommunication~Union, I. 2019.
\newblock ``{World Telecommunication/ICT Indicators Database}.''.

\bibitem[\protect\citeauthoryear{Karlan and Valdivia}{Karlan and
  Valdivia}{2011}]{karlan2011}
Karlan, D., and M.~Valdivia. 2011.
\newblock ``Teaching entrepreneurship: Impact of business training on
  microfinance clients and institutions.''
\newblock {\em Review of Economics and Statistics\/}~{ 93\/}~(2): 510--527.

\bibitem[\protect\citeauthoryear{Maddox, Bradley, Glass, and Filoteo}{Maddox
  et~al.}{2008}]{MaddoxCognition}
Maddox, W.~T., C.~L. Bradley, B.~D. Glass, and J.~V. Filoteo. 2008.
\newblock ``When more is less: Feedback effects in perceptual category
  learning.''
\newblock {\em Cognition\/}~{ 108\/}~(2): 578--589.

\bibitem[\protect\citeauthoryear{McKenzie}{McKenzie}{2020}]{McKenzieTrainingDoesntWork}
McKenzie, D. 2020.
\newblock ``Small business training to improve management practices in
  developing countries: Reassessing the evidence for “training doesn’t
  work”.''
\newblock {\em World Bank Open Knowledge Policy Research Working
  Papers\/}~{9408}.

\bibitem[\protect\citeauthoryear{McKenzie and Woodruff}{McKenzie and
  Woodruff}{2013}]{McKenzieWBRObs}
McKenzie, D., and C.~Woodruff. 2013, 07.
\newblock ``{What Are We Learning from Business Training and Entrepreneurship
  Evaluations around the Developing World?}.''
\newblock {\em The World Bank Research Observer\/}~{ 29\/}~(1): 48--82.

\bibitem[\protect\citeauthoryear{Meager}{Meager}{2022}]{Meager2022a}
Meager, R. 2022, June.
\newblock ``Aggregating distributional treatment effects: A bayesian
  hierarchical analysis of the microcredit literature.''
\newblock {\em American Economic Review\/}~{ 112\/}~(6): 1818--47.

\bibitem[\protect\citeauthoryear{Ministry~of Micro and
  Medium~Enterprises}{Ministry~of Micro and
  Medium~Enterprises, Government of India}{2019}]{MMSMEI(2019)}
Ministry~of Micro, S., and I.~Medium~Enterprises. 2019.
\newblock ``Annual report 2019-20.''.
\newblock Ministry of Micro, Small and Medium Enterprises - Government of
  India.

\bibitem[\protect\citeauthoryear{Mullainathan and Shafir}{Mullainathan and
  Shafir}{2013}]{ms2013}
Mullainathan, S., and E.~Shafir. 2013.
\newblock {\em Scarcity: Why having too little means so much\/} (1st ed.).
\newblock New York: Times Books/Henry Holt and Co.
\newblock ISBN 9780805091239.

\bibitem[\protect\citeauthoryear{van Lieshout and Mehtha}{van Lieshout and
  Mehtha}{2017}]{ILO2017}
van Lieshout, S., and P.~Mehtha. 2017.
\newblock ``The next 15 million start and improve your business global tracer
  study 2011-15.''.
\newblock The International Labour Organization.






\end{thebibliography}

\clearpage




\end{document}